Accountability, Career Incentives, and Local Bureaucrats:...
Transcript of Accountability, Career Incentives, and Local Bureaucrats:...
Accountability, Career Incentives, and LocalBureaucrats: Evidence of A Target-Based Performance
Evaluation System in China
Yvonne Jie Chen∗
National University of SingaporePei Li†
Xiamen UniversityYi Lu‡
National University of Singapore
March 15, 2016
Abstract
This paper studies the effect of a target-based performance evaluation system for
high-level bureaucrats in China. In 2006, the Chinese central government made sul-
phur dioxide (SO2) emissions cuts a main performance evaluation criterion for prefec-
tural city mayors and party secretaries. Using a difference-in-difference framework, we
compare emissions reductions in the so-called Two Control Zone cities, where more
stringent emissions quotas were imposed, with reductions in non-control cities. Our
results suggest that the target-based evaluation system is effective in achieving the
targeted policy output. Adding the emission cuts into local bureaucrats’ performance
evaluations significantly reduces actual SO2 emissions. By exploring a rich set of offi-
cial characteristics, we further analyze the channels through which the policy operates.
We show that the target-based evaluation system changes bureaucrats’ behavior and
induces them to exert more effort on controlling environmental pollution.
∗[email protected]†[email protected]‡[email protected]
1
1 Introduction
Over the past decade, experience with the Millennium Development Goals shows that good
governance supports sustained progress towards economic growth, poverty reduction and
equality. Good governance has also become the center of discussion for the UN post-2015
development agenda (UNESCO [2012]). The quality of state institutions, including the per-
formance of central and local government bureaucracies, holds the key to building effectively
functioning governance in many developing countries (Evans and Rauch [1999], Rauch and
Evans [2000]). How bureaucrats can be motivated to achieve efficient public management
as well as public goods and services delivery is, therefore, an important question for policy-
makers and scholars.
In this paper, we study the effect of a target-based performance evaluation system on
the behavior of local bureaucrats. Career concern theories suggest that top bureaucrats
are largely driven by the outcomes of their mandated tasks (Holmstrom [1982], Holmstrom
and Milgrom [1991], Dewatripont et al. [1999], Alesina and Tabellini [2007, 2008]). Under
this framework, bureaucrats choose their effort levels and their distribution of efforts across
tasks to maximize their signaled capability to the non-public sectors. In addition to career
motives, studies show that bureaucrats are also motivated by peer pressure, social norms,
and bureaucrats’ own “glow of warmth” (Wilson [1989], Maskin and Tirole [2004]).
Various incentive schemes for bureaucrats have been studied in the empirical literature.
Previous studies have generated mixed evidence on the effectiveness of these schemes. The
public administration literature has questioned incentivizing public sector workers with pay-
for-performance. It is only shown to be effective when bundled with other incentives and
proper management practice(Besley [2004], Perry et al. [2009]). Khan et al. [2016] show
that an evaluation system based purely on performance-pay for tax collectors successfully
increased tax revenue by fourteen percent, with no significant political cost. Recently ran-
domized studies have been used to estimate the effectiveness of monetary versus implicit in-
centive schemes such as career advancement(Duflo et al. [2012], Ashraf et al. [2015]). These
2
studies, however, focus heavily on front-line public sector workers rather than bureaucrats
at the management level.
In this paper, we examine the effect of a target-based performance evaluation system
implemented in China in 2006. China has been the fastest-growing major economy over the
past three decades, but growth comes at the cost of heavy environmental pollution1. The
central government has realized the pressing environmental issues since the 1990s and has
gradually shifted its focus from GDP growth to a more sustainable approach that places
a much heavier weight on environmental protection. One of the major policy initiatives
proposed by the central government in the 1990s was the 1998 establishment of the Two
Control Zones policy (for control of sulphur dioxide and acid rain), hereafter TCZ. The
main objective of the TCZ policy was to reduce SO2 emissions, which have long been a
major contributor to China’s ambient air pollution2,3. Based on the TCZ policy guideline,
162 out of 286 prefectural cities were assigned to be TCZ cities and had to follow more
strict environmental regulations than the non-TCZ regions (Figure 1). However, the effect
of the TCZ policy on SO2 emissions reductions was temporary, and emissions continued to
increase after a short dip in 1999-2000 (Figure 2). In order to reach the total SO2 emissions
cut targets (also known as the “green goal”), the central government imposed nationwide
emissions quotas in 2000, with more stringent measures applied to TCZ cities. In 2006,
further administrative regulations were imposed and the emissions quota was brought into
the evaluation system, making local bureaucrats subject to administrative demerit or removal
from office if the quota was not met.1Twelve of the twenty most polluted cities in the world are located in China, and only one percent of the
country’s 560 million city dwellers are breathing air considered safe by the European Union (World Bank[2007]). The total economic cost of China’s air pollution was assessed at 4.1 percent of GDP in 2002 (Wenand Chen [2008]) and the total economic loss of water and air pollution was evaluated at nearly 6.9 percentof GDP in 2003 based on the willingness to pay approach (World Bank [2007]). China consistently ranks atthe bottom quintile for ambient air quality by the World Health Organization.
2The ambient concentration of SO2 in many Chinese cities was among the highest in the world whenChina’s coal consumption reached a peak in 1995. The damage from acid rain in the early 1990s wasestimated to be about 0.7 percent of GDP (World Bank [1997])
3Existing studies have shown that this policy of TCZ affects investment (Lu et al. [2015]), exports (Heringand Poncet [2014]) and health (Tanaka [2015]).
3
Using the implementation of the target-based performance evaluation system imposed
by the central government on TCZ cities and non-TCZ cities, we find that local bureau-
crats exert more effort on SO2 reduction when the emissions quota is built into their perfor-
mance evaluation. Compared to the control group (non-TCZ cities), TCZ prefectural cities
exhibited a similar pattern of reduction in SO2 before 2006; after SO2 reduction became
part of the performance measures in 2006, TCZ prefectural cities experienced much higher
SO2 reductions.
China provides a good setting to study the effectiveness of target-based performance
incentives for bureaucrats for several reasons. First, previous studies suggest that top lo-
cal bureaucrats in China, including provincial and prefectural city government leaders, are
largely motivated by their career concerns (Chen et al. [2005], Li and Zhou [2005], Jia [2014]).
In China, the most important career consideration for local bureaucrats is promotion along
the government hierarchy. Specifically, a city’s China Communist Party (CCP) secretary
aspires to be promoted to the provincial government, while the city mayor seeks promotion
to city CCP secretary or to the provincial government. This career incentive is inline with
bureaucrats in many other countries around world. Findings in China, therefore, help build
general understanding of the impact of a target-based performance evaluation system on
bureaucratic behaviors. Second, at the prefectural city level, city mayors and party sec-
retaries oversee the delivery of public goods and services subject to goals and regulations
set by the central government. They also have a certain degree of autonomy to implement
policies at their own discretion. Hence the delivery of public goods and services varies widely
across China depending on the characteristics and competence of local officials. The regional
variation allows us to identify the policy impact by exploring the rich data on output and
government official characteristics at the prefectural city level.
Our paper adds new empirical evidence on incentivizing bureaucrats with concrete per-
formance targets to the economics literature. Lockwood and Porcelli [2013] show that the
comprehensive performance assessment system in England has increased public service qual-
4
ity but has no significant effect on efficiency overall. Recent studies in Nigeria (Rasul and
Rogger [2015a,b]) suggest that concrete targets have a detrimental effect that leads to dys-
functional responses among bureaucrats. The failure of performance-based incentives is
usually attributed to the multitask nature of bureaucratic jobs and their intrinsic motives
(Dixit [2002], Alesina and Tabellini [2008]). As bureaucrats usually juggle multiple tasks or
multiple dimensions of a task, concert performance targets potentially cause social distor-
tion by requiring bureaucrats to exert effort on less productive outcomes (Rasul and Rogger
[2015b]). Emphasizing output as a performance measure would also likely crowd out the
intrinsic motivations for bureaucrats (Benabou and Tirole [2003]). Our results, however,
suggest that a target-based evaluation system can be effective in motivating bureaucrats to
shift effort towards the task that is more heavily weighted by the evaluators and achieving
the targetted policy output.
Concerns on gaming the target-based performance evaluation system were also brought
up in previous studies (Bevan and Hood [2006], Chen et al. [2012], Gao [2015]. We address
the issue with detailed firm level data from SO2 intensive sectors4. We find that it is much less
likely to establish a new coal fire plant in TCZ cities after the implementation of the target-
based evaluation system. SO2 intensive firms are also more likely to adopt desulfurization
technologies. These evidence suggests that our results are probability not driven by gaming
or data manipulation.
Our paper also adds to the literature on career incentives and occupation selection in
the public sector. Studies on organization theories have shown that employees tend to sort
themselves according to the job attributes and incentives provided. Job incentives would
hence affect the performance outcomes through selection on employees (Roy [1951], Oyer
and Schaefer [2010], Lazear and Oyer [2012],Ashraf et al. [2015]). In analyzing the mecha-
nisms for the new evaluation system, we collected a large amount of information on individual
characteristics, political tenure and job turnover for city mayors and secretaries from various4See section 3.1 for definitions of SO2 intensive sectors and firms
5
Chinese sources. We also extracted words of mentions for environmental related topics and
other government objectives from city governments’ Annual Work of the Government Report
(zhengfu gongzuo baogao) between 2001 and 2013. Our analyses show that SO2 reductions
were not a result of office turnover or sorting of bureaucrats into local offices. This strength-
ened our hypothesis that the observed emissions reduction was the result of changed behavior
and focus of local bureaucrats.
Our finding contributes to a small but growing literature on what motivates local govern-
ments to adopt more stringent environmental policies. Rigor and willingness to implement
environmental regulation at the local government level is usually associated with political
connections, promotion probability and other political concerns (Zheng et al. [2014], Jia
[2014], Kahn et al. [2015]). Our paper shows that the readiness of the center to reward and
punish local officials on the basis of their environmental performance could also motivate
local government officials to reach the green goal. 5
The remainder of the paper is organised as follows. Section 2 provides a review of the
policy background on the TCZ policy and performance evaluation system. We then present
data description, empirical strategy and results in sections 3 and 4.
2 Administrative Enforcement of SO2 Reductions Tar-
gets
SO2 emissions have long been a major contributor to China’s ambient air pollution and
reducing SO2 emissions has been a priority for China’s environmental authorities since the
1990s. Beginning with the 9th Five-Year Plan (FYP) in 1995, China’s central government
began to set limits on total SO2 emissions. In 1998, the State Council instituted an SO2
5Our results complement and echo the literature evaluating the target-based vertical control system (SeeXu [2011] for a review).
6
reduction program, known as the TCZ policy, to limit ambient SO2 pollution and to curb
the growing incidence of acid rain. The TCZ policy encompasses two target policy areas: the
SO2 Control Zone, covering cities in North China, and the Acid Rain Control Zone, covering
cities in South China. Aggregately, the TCZ regions cover 11 percent of China’s territory
(see Figure 1) and are responsible for over 60 percent of China’s total SO2 emissions.6 A
total emissions control policy was incorporated into China’s 10th FYP in 2000 to work in
concert with the TCZ policy7. The goal set by the 10th FYP was to reduce SO2 emissions by
10 percent nationwide and by 20 percent for the TCZ regions, as compared to 2000 levels8.
To force localities to adhere to the national policy, two administrative enforcement mech-
anisms were put into place beginning in 2005. First, starting from Dec 2005, local govern-
ment leaders (mayors and party secretaries) would be held accountable for reaching the
environmental protection goals set by the central government in their administrative region,
including the SO2 reduction goal9. The top-down target-based responsibility scheme ensures
that local officials are tied to satisfying higher-level mandates for career advancement and
legitimacy10. Second, the central government announced in 2007 that the SO2 emissions
reduction targets to be attained by the end of 2010 would be incorporated into the responsi-
bility contracts signed with upper-level governments. SO2 emissions reductions hence became
one of the most important factor in the assessment of local cadres. Local government leaders
who fail to meet the annual SO2 emissions reduction targets would be penalized with an6Relevant Chinese document for the TCZ policy is available at following government website: http:
//www.mep.gov.cn/gkml/zj/wj/200910/t20091022_172231.htm7Relevant Chinese document for the 10th FYP guideline for TCZ policy is available at following govern-
ment website: http://www.ynepb.gov.cn/xxgk/read.aspx?newsid=110128A series of policies were implemented in TCZ regions in order to meet the emissions reduction targets,
including the closure of high SO2 concentration coal mines, more control of SO2 emissions from coal-firedpower plants and industrial boilers, and the introduction of more energy-saving and low pollution technologies(Cao et al. [2009]). More generous fiscal policies were also implemented in TCZ regions, including subsidiesfor energy saving projects and the establishment of an SO2 quota trading system. Higher pollution fees werealso imposed in some regions.
9Relevant Chinese document is available at following government website: http://www.gov.cn/zwgk/2005-12/13/content_125680.htm
10In the meantime, the central government banned new construction projects in regions with unsatisfac-tory SO2 emissions reduction, in an effort to align incentives for economic growth with local governments’achievement of the green goal.
7
administrative demerit or removal from office11. The administrative enforcement ensures the
effectiveness of the TCZ policy across the chessboard of China’s territorial administration
and further ties local cadres to their fulfillment of the target SO2 emissions reductions.
Before the administrative enforcement policies were imposed, the green goal set in China’s
10th FYP (in 2000) was not strictly implemented. By the end of 2005, SO2 emissions not
only had not reduced, but had in fact increased by approximately 28 percent from 2000
levels (see Figure 2). After 2005, the consistent emphasis by the central government on
the need to reduce SO2 emissions has made it incrementally more difficult for both local
governments and polluters to drag their feet in carrying out the central government’s SO2
emissions reduction policy. By 2010, the SO2 emissions reduction targets was finally met
and an overall 14.29 percent reduction achieved nationwide (see Figure 2).
3 Estimation Strategy
3.1 Data and Variables
Data for the empirical analyses presented in this section are collected by the authors from
various official statistical publications and public databases. Using these data sources we
construct a data set containing environmental, socioeconomic and meteorological conditions
of each city spanning the years from 2001 to 2013. Summary statistics are presented in Table
1.
Pollution Data
The industrial SO2 emissions and wastewater discharge data are collected from the an-
nual statistical yearbooks of the Chinese provinces, supplemented by China City Statistical11Relevant Chinese document is available at following government website: http://www.gov.cn/jrzg/
2007-06/03/content_634545.htm
8
Yearbooks 2004-2014 and China Environmental Yearbooks 2002-201412. We use emission data
between 2001 and 2013 for analyses presented in this paper. Data before 2001 were subject
to significant missing values hence were not included in the study. The city-level annual SO2
emissions are calculated by summing up emissions from all reporting firms in a given year.
For each reporting firm, the State Environmental Protection Agency (SEPA) records its SO2
emissions through direct monitoring. If direct monitoring is not feasible, SEPA would use
reverse engineering measures computed from the industrial waste reported by the firm. The
list of reporting firms is determined by SEPA and is subject to minor revisions each year. The
city-level coal-fired generation capacity data are aggregated from the plant-level data docu-
mented in the Compilation of Statistical Materials of Electric Power Industry 2001-2010,13
and the SEPA’s annual communiques on the shutdown of small thermal power plants.14 The
PM2.5 data are extracted from the grid data of Global Annual PM2.5 Grids from MODIS
and MISR Aerosol Optical Depth, which provides a continous surface of concentrations of
particulate matter of 2.5 micrometers or smaller15.
Other City-Level Variables
The city slope and elevation are extracted from the Shuttle Radar Topographic Mission
90m Digital Elevation Model data. We collect city wind speed, temperature and precipitation
data from the China Meteorological Data Sharing Service System (http://cdc.cma.gov.cn/).
Soil pH data are extracted from the grid data of the Global Dataset of Derived Soil Properties
at 0.5 degree by 0.5 degree. Data on total employees in the “dirty sectors” are from the China
Establishment Census 1996 . We aggregate the firm-level data by sector and city. A dirty
sector is defined by matching the 2-digit industry classification code to that of major SO2
pollution sectors identified by the Report of China Pollution Source Census 2007. As the
only pollution source census in China up to this date, the 2007 census examined 5.92 million12According to the China National Pollution Census 2008, industrial activity is responsible for 91.4 percent
of total SO2 emissions. The remaining part is contributed by domestic activity (8.6 percent).13This dataset covers all thermal power plants with installed capacity of 6,000 KW and above.14SEPA has stopped publicly publishing the list of small thermal power plants shutdown annually in 2011.15The raster grids have a grid cell resolution of 30 arc-minutes and cover the world from 70°N to 60°S
latitude.
9
pollution sources across all provinces in China. The final report identified six industries (as
shown in Table 2) that rank the highest in terms of SO2 emissions share in total industrial
SO2 emissions. These six industries comprise 88.5% of total SO2 industrial emissions. The
fatality data of workplace accidents are collected from the official website of the China State
Administrative of Work Safety (http://www.chinasafety.gov.cn). We aggregate the daily
data by year and city.
Characteristics of City Government Officials
We collect detailed demographic data including age, education level and length of service
at current position. The detailed biographical data are compiled manually by the authors
from various publicly accessible sources, including the online biographical database China
Vitae (http://www.chinavitae.com/).
Mentions of Environmental Protection
We conduct a thorough content analysis of the city governments’ Annual Work of the
Government Report between 2001 and 2013. The Annual Work of the Government Report is
usually presented by the city mayor at the National People’s Congress early in the year. The
report sets the goals and objectives for city governments’ work for the upcoming year and
is meant to be supervised by the general population. Each report outlines the target GDP
growth rate for the year as well as other policy objectives for the city government. The reports
are not available in any existing database or statistical year books. We collect information
by manually searching more than 3,700 reports via city governments’ websites and local
newspapers. Our paper is, to our knowledge, the first attempt to use information extracted
from the Annual Work of the Government Report for empirical analysis. Specifically, we
search for the words “environmental protection” (huan bao or huanjing baohu in Chinese)
and “emissions reductions” (jian pai in Chinese), and calculate the ratio of these mentions
to the total number of words in the report. We use the ratio as a measure of effort that a
city government leader places on environmental issues.
10
3.2 Estimation Framework
To investigate whether the performance-based evaluation system affects government officials’
behavior in implementing and enforcing environmental regulations, we compare the emissions
of SO2 in TCZ cities (those cities with strict environmental regulations) before and after 2006
(the year that the performance-based evaluation system was adopted) with that of non-TCZ
cities during the same period, or a difference-in-differences (DD) estimation. Specifically,
our regression equation is
LnSO2ct = βTCZc ∗ Post05t + δc + γt + Zc ∗ f(t) + εct, (1)
where LnSO2ct is the logarithm of SO2 emission in city c at yeart; TCZc equals 1 if city c
was designated as a TCZ city in 1998 and 0 otherwise; Post05t takes a value of 1 if t > 2005
and 0 otherwise; δc is city fixed effect, capturing all time-invariant differences across cities; γt
is year fixed effect, capturing all yearly factors that are common to cities such as macro-level
shocks; Zc denotes the determinants of TCZ selection measured in the pre-treatment period
and f(t) is a third-order polynomial function, both explained later; and εct is the error term.
To accommodate potential heteroskadesticity and serial correlation, we cluster the standard
errors at the city, following the suggestion by Bertrand et al. [2004].
We expect β to be negative if the performance-based evaluation system is effective. This
is because TCZ cities were imposed larger targeted reductions in SO2 in 2000 and the new
evaluation system was introduced in 2006. However, the unbiased estimation of β requires
that SO2 emissions in TCZ cities would have followed the same time trend as that in non-
TCZ cities after 2005 if a performance-based evaluation system had not been imposed by
the central government in 2006. A primary threat to our identifying assumption is that
cities were not randomly selected into TCZ and non-TCZ groups in 1998. This non-random
selection implies that TCZ and non-TCZ cities could have been systematically different before
1998. If these unobserved differences changed their values in 2006, our estimate would be
11
biased. To address this concern, we follow an approach used by Gentzkow [2006], in which
key determinants of the TCZ selection are first identified and then post-treatment variations
in the outcome variable generated by these key determinants are flexibly controlled in the
regression. The premise of this approach is that conditional on these key determinants, the
central government did not select TCZ cities on remaining unobservables with a perspective
that there would be a new performance-based evaluation system eight years later and these
remaining unobservables would behave differently between TCZ and non-TCZ cities at that
time.
Given that the designation of TCZ cities was initiated by the central government in the
mid-1990s, we look at government policy documents to understand the factors that shaped
the government’s decision-making processes on the TCZ selection. Specifically, according
to the Air Pollution Prevention and Control Law (APPCL) amended in 1995, based on
conditions in the atmosphere, terrain and soil, cities could be designated as either an SO2
Control Zone or an Acid Rain Control Zone. In addition, the 1998 TCZ policy classified
those cities that were already seriously polluted by SO2 or acid rain as TCZ cities. We
collect two sets of variables as our TCZ selection variables. The first set of variables reflect
conditions of the natural environment based on the 1995 APPCL amendment: Roughness
(the standard deviation of slope), Elevation (average elevation in km), Wind Speed (annual
average wind speed in 1990-1995), Precipitation (annual average precipitation in 1990-1995),
Soil pH (average pH level of topsoil), Coldness (percentage of days with temperature of 5oor
below in 1990-1995). The second set of variables reflect pre-policy SO2 pollution levels,
which include Coldness (percentage of days with temperature of 5oor below in 1990-1995)
and the total employment in “dirty sectors” in 199616. To flexibly control for trends in
SO2 emissions generated by these selection variables over time, we interact them with a16Ideally we should include pre-policy level SO2 level as a determinant for TCZ designation for each
city but unfortunately the data are not available. We therefore use pollutants that are known to be highlycorrelated with SO2 and employees in the dirty sector as a proxy for baseline SO2. When the daily outdoortemperature drops to 5°C or below for a few days, the north part of China enters the heating period. As aresult, coal combustion in boilers is associated with the release of air pollutants that are highly correlatedwith SO2.
12
third-order polynomial function of time Zc ∗ f(t).
As further checks on our identifying assumption in equation 1, we conduct several sen-
sitivity exercises, including a graphic verification of whether TCZ and non-TCZ cities had
similar trends in SO2 emissions before the treatment, a placebo test with random assignment
of TCZ status and timing of the adoption of the performance-based evaluation system, and
falsification tests with other pollutants that the policies did not address. For details on these
robustness checks, see Section 4.2.
4 Empirical Findings
4.1 Main Results
Table 3 reports main results from equation 1. Column 1 presents results from a simple DD
estimation with only city and year fixed effects included. The interaction between TCZc
and Post05t is negative and statistically significant, suggesting that SO2 emissions fell more
significantly in TCZ cities than in non-TCZ cities after 2005. Given that the TCZ policy
imposed stricter regulations on SO2 emissions in TCZ cities than in non-TCZ cities, and that
a performance-based evaluation system was imposed in 2006, this result implies that the new
evaluation system has disciplined or incentivized bureaucrats to enforce the environmental
regulations.
To alleviate the concern that our estimate is biased due to the non-random selection of
TCZ cities in 1998, we include interactions between determinants of TCZ selection and a
third-order polynomial function of time in column 2. We continue to find a negative and
statistically significant estimate of TCZc∗Post05t, despite a drop of around 40 percent in the
magnitude. In column 3, we further include, as a control variable, an interaction between
TCZc and Y ear05t to investigate whether bureaucrats anticipated the new performance-
13
based evaluation system in 2006 and hence changed their behavior before the adoption
(known as the Ashenfelter dip). We find a statistically insignificant estimate of TCZc ∗
Y ear05t with barely any change in magnitude, suggesting no Ashenfelter dip in our research
setting.
As the new performance-based evaluation system consists of two measures implemented
respectively in 2005 and 2007, we divide our regressor of interest into two, TCZc∗Y ear06 07t
and TCZc ∗ Post2007, to incorporate the effects from the two post-treatment periods. As
shown in column 4, both of the two regressors are negative and statistically significant,
with the latter having about 78 percent larger magnitude, suggesting that both of the two
measures in the new performance-based evaluation system matter.
4.2 Robustness Checks
Our aforementioned estimates hinge on the identifying assumption that conditional on the
determinants for TCZ selection, the central government did not select TCZ cities on the re-
maining factors with the expectation that eight years later it would pass a new performance-
based evaluation system and these remaining factors would change the trends of SO2 emis-
sions between TCZ and non-TCZ cities differently after the new evaluation system. To
further verify that our identifying assumption sustains, we report in this section a battery
of robustness checks.
Similar time trends before the treatment. A necessary condition for satisfying our identi-
fying assumption is that TCZ and non-TCZ cities have similar time trends in SO2 emissions
before the treatment. Specifically, if similar time trends of SO2 emissions existed for a long
period before the treatment and suddenly diverged right after the treatment, it may indicate
that the differential trends may be caused by the treatment. Figure 3 plots the time trend
of differences in SO2 emissions between TCZ and non-TCZ cities over 2002-2013 relative
14
to the difference in 2001. Each dot in Figure 3 estimates the additional gap in emissions
between TCZ and non-TCZ cities using the gap in 2001 as the baseline17. Indeed, we find
that the differences are close to zero from 2001 to 2005, suggesting a common trend between
TCZ and non-TCZ cities for the five years before the treatment. Meanwhile, the difference
dropped to negative immediately after the adoption of the treatment in 2006 and continued
to be negative with increasing magnitude. These results confirm that the treatment in 2005
generated the differential trends in SO2 emissions between TCZ and non-TCZ cities.
A placebo test with randomization of the treatment. As a second robustness check, we
randomly assign the TCZ status to cities TCZfalsec and the timing of the adoption of the
new performance-based evaluation system Postfalset , and then construct a new regressor of
interest TCZfalsec ∗Postfalset . Given this random data generating process, a necessary condi-
tion for satisfying our identifying assumption is that TCZfalsec ∗Postfalset shall produce zero
effect on SO2emissions; otherwise, it indicates that our estimation equation is misspecified.
To increase the power of this test, we conduct the random assignment 1,000 times. The
distribution of 1,000 estimated coefficients is plotted in Figure 4. It is centered around zero
and only 1.1 percent of the estimates are more negative than our true estimate in column 2
of Table 3. These results suggest that there likely were not substantial omitted variables in
our specification18.
Alternative pollutant measures. We use PM2.5 as an alternative measure of SO2 emis-
sions. PM2.5 data have been used extensively in the environment literature and are shown
to be a reliable measure of air quality. PM2.5 particles can be directly emitted or pro-
duced from emitted precursors. Pollutants such as SO2, nitriate (NOx) and volatile organic
compounds (VOCs) can react in the atmosphere to produce PM2.5. Although changes in
PM2.5 can be a complicated process, PM2.5 can be used as a good alternative measure for
SO2 emissions. The estimation result for the policy effect on PM2.5 particles is reported in
column 5 of Table (3). The estimated coefficient is negative and statistically significant. A17All other control variables are the same as column 2 in Table 3.18See Chetty et al. [2009] and La Ferrara et al. [2012] for review and application of the method
15
decrease in PM2.5 as a result of SO2 control policy in absence of regulations for the other
reactors supports our hypothesis that the policy has effectively reduced SO2 emissions.
Analyses with untargeted pollutants. Our estimates would be biased if, in 2006, along
with the change in the evaluation system, there were other differential changes between TCZ
and non-TCZ cities such as industrial structure and technology upgrading that affected
environmental conditions. However, the TCZ policy exclusively targeted SO2 emissions and
did not address other pollution measures, such as wasterwater. Hence, if our aforementioned
estimates are mainly due to an evaluation system that made bureaucrats more concerned
about the enforcement of the TCZ policy, we shall not spot sizable differentials among
alternative pollution measures between TCZ and non-TCZ cities after 2006. The estimation
result for the policy effect on wastewater is reported in column 6 of Table 3. The estimated
coefficient is small in magnitude and insignificant. The result lends further support to the
satisfaction of our identifying assumption.
4.3 Effect of the Administrative Enforcement on GDP Growth and IndustrialSectors
First, we estimate the effect of the performance evaluation system on GDP growth rate
and target GDP rate at the city level. Historically, GDP growth has been the indicator
of competency of local government officials and has been used as a basis for promotion
(Jia [2014], Xu [2011]). As outlined in Appendix A, when there is a trade-off between
multiple tasks, bureaucrats will exert more effort on the task that’s more heavily weighted
by their supervisors. Hence, the shift of focus to environmental protection may lead to a
reduced emphasis on economic growth by the city government. To this end, we examine the
target GDP growth rate mentioned in the city governments’ Annual Work of the Government
Report, in both absolute terms and relative terms (that is, relative to the target GDP growth
rate by the provincial government). Estimation results are reported in columns 1 and 2 of
Table 4. Consistently, we find that after 2005, TCZ city had lower GDP growth rates, as
16
shown in columns 3 in Table 4, indicating a shift of focus away from economic growth in
response to the new evaluation system.
In order to examine the policy effect on industrial sectors, we first examine the proba-
bility of coal fire plants entry and exit at the city level. To control SO2 emissions, the TCZ
policy restricted the opening of coal fire plants and demanded the closing of outdated coal
fire plants in the TCZ cities. Hence, if our aforementioned estimates are caused by a change
in enforcement of the TCZ policies due to the new evaluation system in 2006, we shall find
consistent patterns of opening and closing of coal fire plants, specifically, less openings and
more closings in TCZ cities after 2005. Indeed, the regression results in columns 1 and 2
of Table 5 confirm these results: TCZc ∗ Post05t is negative and statistically significant in
the regression of plant openings, and positive and statistically significant in the regression
of plant closings. We also check the probability of installing desulfurization equipment at
firm level. We expect that the performance evaluation would lead to a higher probability of
installing desulfurization equipment among TCZ cities as an enforcement measure taken by
local bureaucrats to reduce SO2 emissions. Results in column 3 of Table 5 confirm our hy-
pothesis. The coefficient for TCZc ∗Post05t is positive and statistically significant, meaning
that firms in TCZ cities are indeed more likely to install these environmental protection tech-
nologies. These results provide micro-evidence that the target-based performance evaluation
system has led to sectoral shifts and technology changes at the plant level.
4.4 Mechanisms and Alternative Explanations
We have documented that SO2 emissions dropped in TCZ cities relative to that of non-TCZ
cities after 2005. The change in outputs can be explained by theories of career concerns as
outlined in Appendix A. Motivated by the theoretical works of Holmstrom [1989] and Alesina
and Tabellini [2007], we hypothesize that bureaucrats exert more effort on controlling SO2
emissions when the cut quota is built into the evaluation system.
17
We first provide a content analysis of the city governments’ Annual Work of the Gov-
ernment Report. We use this to examine the emphasis a city government places on envi-
ronmental issues in the coming year. Specifically, we search for the words “environmental
protection” (huan bao or huanjing baohu in Chinese) and “emissions reduction” (jian pai in
Chinese), and calculate the ratio of the number of these mentions to be the total number of
words in the report. We then examine whether the government reports in TCZ cities had
more coverage of environmental protection than in non-TCZ cities after 2005. Regression
results are reported in column 4 of Table 4. We find that TCZc ∗Post05t is positive and sta-
tistically significant, suggesting that TCZ city governments became more concerned about
environmental protection after 2005 than non-TCZ city governments.
We also examine the effect of random onset of fatal accidents in a city on SO2 emissions
levels. The new performance-based evaluation system reduces the probability of promotion
if the bureaucrats did not meet the annual SO2 emissions cut target. However, other poli-
cies also affect a bureaucrat’s likelihood of promotion. In particular, a provision issued by
the central government in 2001 declared that “when an extremely fatal accident happens,
the local government officials would be penalized based on the severity of the accident, ei-
ther through demotion or dismissal, or through criminal procedures if dereliction of duty
is involved”19. Hence, in the case of a fatal accident, officials’ promotion probability would
be negatively affected20. In the event of a fatal accident, the new performance evaluation
system would incentivize bureaucrats to exert even more effort towards SO2 emissions cuts
so that they can compensate for the fatal accident21. Following this reasoning, we examine
whether the occurrence and severity of an accident further increase the gap in the reduction19The Chinese document can be found on the website of China government at: http://www.sxaj.gov.
cn/flfg/fg/gjfl/200905/881148.html. A workplace accident is regarded as very serious if there are morethan 10 fatalities.
20Fisman and Wang [2015] show that firms with political connections do not receive favorable treatmentwhen fatal accidents happen and connections-mortality relationship is reduced in provinces where localofficials are subject to certain safety targets. Their evidence suggests that it is difficult for local officials toavoid regulatory responses when fatal accidents happen.
21Only 3 out of 2499 city secretaries/mayors had been demoted because of accidents during our targetedperiod. As a result, the incidence of fatal accidents merely reduces the likelihood of promotion; it does notcompletely destroy the possibility of promotion.
18
of SO2 emissions between TCZ and non-TCZ cities after 2005 in Table 6. The triple in-
teraction term, TCZc ∗ Post05t ∗ Fatal−Accidentct, is negative and statistically significant,
regardless of Fatal−Accidentct is measured by a dummy variable indicating with or without
fatal accidents or a continuous variable capturing the death toll. Our results indicate that
local bureaucrats treat SO2 reductions as a coping mechanism when an unforseenable event
happens and exert more effort on environmental issues in order to make up for what has
happened.
An alternative explanation for the observed reduction in SO2 is occupation selection.
Many previous studies suggest that incentive packages and attributes of a job sort individuals
into the “right” type of jobs (Ashraf et al. [2014, 2015], Roy [1951]). Hence it is possible that
more environmentally conscious bureaucrats were put into office in TCZ cities after 2005 as
compared to non-TCZ cities.
In China, the most important bureaucrat at the city level is the secretary of the CCP, who
is in charge of the city’s major decisions; the second most important bureaucrat is the city
mayor, who is responsible for the daily operation of the government. To understand whether
our findings are driven by occupation selection of bureaucrats with different environmental
orientations across TCZ and non-TCZ cities, we conduct two sets of exercises.
First, we check whether the pattern of length of appointment and frequency of office
turnover diverge between TCZ and non-TCZ cities after 2005. Specifically, we examine
the probability of having a new CCP secretary and city mayor, the term length of the
incumbent CCP secretary and city mayor, and the promotion of CCP secretary and city
mayor. Regression results are reported in Table 7, with columns 1-3 for city mayors and
columns 4-6 for CCP secretaries. Consistently, we find all the estimates are statistically
insignificant and small in magnitude. These results indicate that TCZ and non-TCZ cities
had similar patterns of CCP secretary/city mayor turnover, term length of the incumbent
CCP secretary/city mayor, and promotion probability of CCP secretary/city mayor after
2005.
19
Second, we examine various individual characteristics of city CCP secretaries and city
mayors to check whether TCZ and non-TCZ cities attracted different types of bureaucrats
after 2005. Specifically, we look at age, education level, and whether the bureaucrats are
working in their hometown in Table 8. We also look at various indicators of bureaucrats’
working experience, including service length in the concerned city, previous experience in a
firm, previous experience in upper-level government, and promotion from lower-level county
government in Table 9. Across all the regressions, we consistently find that the coefficients of
TCZc ∗Post05t are statistically insignificant and small in magnitude. These results indicate
that TCZ and non-TCZ cities have similar governing bureaucrats after 2005.
In summary, while we are unable to directly check whether TCZ and non-TCZ cities
have governing bureaucrats with different preferences or competence levels for implementing
environment regulations after 2005, our analyses show that these two groups of cities have
similar types of bureaucrats and no difference in office turnover and length of service. This
helps dispel the argument that our results are driven by sorting of more environmentally
concious bureaucrats into positions in TCZ cities than non-TCZ cities in the post-2005
period. Our tests also do not suggest that the observed results were due to divergent data
manipulation patterns between TCZ and non-TCZ cities.
5 Conclusion
In this paper we examine the effect of a target-based performance evaluation system on SO2
emissions reduction. We find that when SO2 emissions quota were built into local bureau-
crats’ evaluation system, SO2 emissions were significantly reduced. Our analysis suggests
that the new evaluation system induced local bureaucrats to exert more effort on environ-
mental issues. Using a comprehensive dataset that we constructed, we observe a clear shift
of effort and focus from GDP output to environmental protection among these local bu-
20
reaucrats. Our finding does not suggest that the emissions reductions were a result of office
turn-over or sorting.
Our finding adds to the empirical debate on the effectiveness of target-based performance
evaluation systems for public sector workers, particularly for management-level bureaucrats.
As the earlier literature has highlighted, the addition of a concrete performance target into
an evaluation system might affect the overall comprehensive output as bureaucrats will divert
their attention to the output that’s more heavily weighted by their supervisors. It might
also crowd out bureaucrats’ intrinsic motivation. Our analysis shows a clear pattern of local
bureaucrats making a conscious trade-off between GDP growth and SO2 emissions when
emissions targets are brought into the picture.
Our results indicate that setting a concrete target-based performance evaluation system
for environmental policies can be effective in ensuring that bureaucrats work to curb pollution
and promote environmental protection. The effect is immediate and the impact is significant.
Developing countries facing similar pressing environmental issues might want to consider the
implementation of such a policy.
21
Appendix AA Conceptual Framework
In this section, we present a conceptual framework that illustrates the behavior of bu-
reaucrats in response to a target-based evalaution system. Following the seminal work by
Dewatripont et al. [1999] and Alesina and Tabellini [2007], bureaucrats are partly motivated
by their “career concerns”. In China, career concerns for bureaucrats include moving up the
hierarchical leadership ladder or moving to a state-owned enterprise.
Following Alesina and Tabellini [2007], suppose that a bureaucrat focuses on two policy
outcomes, y1 and y2. To achieve each outcome, the bureaucrat has to exert effort level a1 and
a2, respectively. Let the cost of exerting effort be C(a1, a2) and the cost function be strictly
increasing and convex. This cost is assumed to be public information. The bureaucrat also
cares about the evaluation by her supervisor as it matters for her career concerns. The
policy output for each period is therefore y1 = ηi +a1 + ε1 and y2 = ηi +a2 + ε2, respectively,
where ηi is the ability for a typtical bureaucrat i drawn from a distribution with mean ηi
and variance σηi. ε1, ε2 are random draws from a i.i.d. distribution. Suppose the aggregate
performance of the bureaucrat is assessed by x = φ× y1 + (1− φ)× y2, 0 < φ < 1, where φ
is the weight assigned by the supervisor to each implementation output. x therefore is the
performance measure of the bureaucrat according to some stated goals.
Let δ be the market value of the signaled ability, then return for the bureaucrat can
be written as RB(a1,a2) = δE(ηi|x)22. The goal for the bureaucrat is to choose her effort
level so that RB(a1, a2)−C(a1, a2) is maximized. First order conditions of the maximization
problem yield
δφ = C1(a1,a2)
δ(1− φ) = C2(a1,a2)22Note that x = φy1+(1−φ)y2 = φ(ηi+a1+ε1)+(1−φ)(ηi+a2+ε2) = ηi+φa1+φε1+a2(1−φ)+ε2(1−φ).
Therefore E(ηi|x) = x−φa1− (1−φ)a2−E[φε1 + (1−φ)ε2|x] and can be written as a function of a1and a2.
22
where C1, C2 is the partial derivative of the cost function with respect to effort a1, a2.
It is easy to show that φ(1−φ) = C1
C2. This means that a necessary condition for optimization
is that the ratio of the marginal cost of effort level between the two tasks is equal to the
relative weight assigned by the evaluator. Given the convex cost structure, a higher weight
leads to increased effort in a particular task.
23
ReferencesA. Alesina and G. Tabellini. Bureaucrats or politicians? part i: A single policy task.
American Economic Review, 97:169–179, 2007.
A. Alesina and G. Tabellini. Bureaucrats or politicians? part ii: Multiple policy tasks.Journal of Public Economics, 92:426–447, 2008.
N. Ashraf, O. Bandiera, and B. K. Jack. No margin, no mission? a field experiment onincentives for public service delivery. Journal of Public Economics, 120:1–17, 2014.
N. Ashraf, O. Bandiera, and S. S. Lee. Do-gooders and go-getters: Career incentives, selec-tion, and performance in public service delivery. Working Paper, 2015.
R. Benabou and J. Tirole. Intrinsic and extrinsic motivation. Review of Economic Studies,70:489–520, 2003.
M. Bertrand, E. Duflo, and S. Mullainathan. How much should we trust differences-in-differences estimates? Quarterly Journal of Economics, 119(1):249–275, 2004.
T. Besley. Paying politicians: Theory and evidence. Journal of the European EconomicAssociation, 2:193–215, 2004.
G. Bevan and C. Hood. What’s measured is what matters: Targets and gaming in the englishpublic health care system. Public Administration, 84(3):517–538, 2006.
J. Cao, R. Garbaccio, and M. S. Ho. China’s 11th five-year plan and the environment:Reducing so2 emissions. Review of Environmental Economics and Policy, 3:231–250, 2009.
Y. Chen, H. B. Li, and L. A. Zhou. Relative performance evaluation and the turnover ofprovincial leaders in china. Economics Letters, 88:421–425, 2005.
Y. Y. Chen, G. Z. Jin, N. Kumar, and G. Shi. Gaming in air pollution data? lessons fromchina. The B.E. Journal of Economic Analysis & Policy, 12(3):1–43, 2012.
R. Chetty, A. Looney, and K. Kroft. Salience and taxation: Theory and evidence. AmericanEconomic Review, 99(4):1145–1177, 2009.
M. Dewatripont, I. Jewitt, and J. Tirole. The economics of career concerns, part ii: Applica-tion to missions and accountability of government agencies. Review of Economic Studies,66:199–217, 1999.
A. Dixit. Incentives and organizations in the public sector: An interpretative review. Journalof Human Resources, 37(4):696–727, 2002.
E. Duflo, R. Hanna, and S. P. Ryan. Incentives work: Getting teachers to come to school.American Economic Review, 102(4):1247–1278, 2012.
24
P. B. Evans and J. E. Rauch. Bureaucracy and growth: A cross-national analysis of theeffects of ”weberian” state structures on economic growth. American Sociological Review,64(5):748–765, 1999.
R. Fisman and Y. X. Wang. The mortality cost of political connections. Review of EconomicStudies, Forthcoming, 2015.
J. Gao. Pernicious manipulation of performance measures in china’s cadre evaluation system.China Quarterly, 223:618–637, 2015.
M. Gentzkow. Television and voter turnout. Quarterly Journal of Economics, 121:931–972,2006.
L. Hering and S. Poncet. Environmental policy and exports: Evidence from chinese cities.Journal of Environmental Economics and Management, 68:296–318, 2014.
B. Holmstrom. Design of incentive schemes and the new soviet incentive model. EuropeanEconomic Review, 17(2):127–148, 1982.
B. Holmstrom. Agency costs and innovation. Journal of Economic Behavior & Organization,12(3):305–327, 1989.
B. Holmstrom and P. Milgrom. Multitask principal-agent analyses: Incentive contracts, assetownership, and job design. Journal of Law, Economics, & Organization, 7:24–52, 1991.
R. X. Jia. Pollution for promotion. Working Paper, 2014.
M. E. Kahn, P. Li, , and D. X. Zhao. Water pollution progress at borders: The role ofchanges in china’s political promotion incentives. American Economic Journal: EconomicPolicy, 7(4):223–42, 2015.
A. Q. Khan, A. I. Khwaja, and B. A. Olken. Tax farming redux: Experimental evidence onperformance pay for tax collectors. Quarterly Journal of Economics, Forthcoming, 2016.
E. La Ferrara, E. A. Chong, and S. Duryea. Soap operas and fertility: Evidence from brazil.American Economic Journal: Applied Economics, 4(4):1–31, 2012.
E. P. Lazear and P. Oyer. Personnel economics. In R. Gibbons and J. Roberts, editors,Handbook of Organizational Economics. Princeton University Press, 2012. NBER WorkingPaper No. 13480.
H. B. Li and L. A. Zhou. Political turnover and economic performance: the incentive roleof personnel control in china. Journal of Public Economics, 89:1743–1762, 2005.
B. Lockwood and F. Porcelli. Incentive schemes for local government: Theory and evidencefrom comprehensive performance assessment in england. American Economic Journal:Economic Policy, 5(3):254–286, 2013.
Y. Lu, M. Q. Wu, and L. H. Yu. Does environmental regulation drive away inbound foreigndirect investment? evidence from a quasi-natural experiment in china. Mimeo, 2015.
25
E. Maskin and J. Tirole. The politician and the judge: Accountability in government.American Economic Review, 94(4):1034–1054, 2004.
P. Oyer and S. Schaefer. Personnel economics: Hiring and incentives. In Handbook of LaborEconomics, volume 4. Elsevier, 2010. NBER Working Paper No. 15977.
J. L. Perry, T. Engbers, and S. Y. Jun. Back to the future? performance-related pay,empirical research, and the perils of persistence. Public Administration Review, 69(1):39–51, 2009.
I. Rasul and D. Rogger. The impact of ethnic diversity in bureaucracies: Evidence from thenigerian civil service. American Economic Review: Papers & Proceedings, 105(5):457–461,2015a.
I. Rasul and D. Rogger. Management of bureaucrats and public service delivery: Evidencefrom the nigerian civil service. Working Paper, 2015b.
J. E. Rauch and P. B. Evans. Bureaucratic structure and bureaucratic performance in lessdeveloped countries. Journal of Public Economics, 75:49–71, 2000.
A. D. Roy. Some thoughts on the distribution of earnings. Oxford Economic Papers, 3(2):135–146, 1951.
S. Tanaka. Environmental regulations on air pollution in china and their impact on infantmortality. Journal of Health Economics, 42:90–103, 2015.
UNESCO. Thematic think piece: Governance and development. Technical report, UNDESA,UNDP, UNESCO, 2012.
Z. G. Wen and J. N. Chen. A cost-benefit analysis for the economic growth in china.Ecological Economics, 65(2):356–366, 2008.
J. Q. Wilson. Bureaucracy: What Government Agencies Do and Why They Do It. NewYork: Basic Books, Inc., New York, 1989.
World Bank. Cost of pollution in china. Technical report, Washington, DC: World Bank,Washington, DC, 1997.
World Bank. Cost of pollution in china. Technical report, Washington, DC: World Bank,East Asia and Pacific Region, 2007.
C. G. Xu. The fundamental institutions of china’s reforms and development. Journal ofEconomic Literature, 49:1076–1151, 2011.
S. Q. Zheng, M. E. Kahn, W. Z. Sun, and D. L. Luo. Incentives for china’s urban may-ors to mitigate pollution externalities: The role of the central government and publicenvironmentalism. Regional Science and Urban Economics, 47:61–71, 2014.
26
Table 1: Summary StatisticsKey Variables Definition Mean S.D.TCZ =1 if a city is a TCZ city; =0 otherwise 0.566 0.496SO2 Total industrial SO2 emissions (104 tons) 6.064 6.078PM2.5 Concentrations of PM2.5 (micrograms/m3) 31.661 14.112Wastewater Total industrial waste water discharges (107 tons) 7.734 10.44CoalFirePlant CapacityEntry New fossil-fuel power plant electricity generation capacity (104KW) 45.050 176.998Exit Closed fossil-fuel power plant electricity generation capacity (104 KW) 0.204 0.760Fatal AccidentHappened =1 if an extremely fatal accident happened; =0 otherwise 0.169 0.370Death Toll Total fatality of the extremely fatal accidents (100 persons) 0.046 0.161Gov EcoIndex Count of ”green” keywords in the government work report
divided by total word count of the report in this year (×103) 0.399 0.287GDP Real Real GDP Growth (%) 12.820 3.430GDPTarget GDP growth target (%) 12.180 2.850GDPTarget Gap City GDP growth target minus provincial GDP growth target 2.251 2.269
City mayer characteristicsAge Age 50.460 4.153Education =1 if the with a master or higher degree; =0 otherwise 0.687 0.467Hometown =1 if born in this city; =0 otherwise 0.090 0.289New =1 if in the first yea of his tenure; =0 otherwise 0.296 0.456Tenure Length of tenure in the current position 2.615 1.589Service Length Total service years in this city 6.912 9.542Promotion =1 if promoted this year; =0 otherwise 0.188 0.391As a leader in aFirm =1 if worked as the CEO in a big enterprise; =0 otherwise 0.197 0.398Up-tier government =1 if worked in a provincial/central government; =0 otherwise 0.518 0.500County =1 if worked as a county governor/secretary; =0 otherwise 0.428 0.495City secretary characteristicsAge Age 52.600 4.056Education =1 if the with a master or higher degree; =0 otherwise 0.666 0.472Hometown =1 if born in this city; =0 otherwise 0.053 0.225New =1 if in the first yea of his tenure; =0 otherwise 0.266 0.442Tenure Length of tenure in the current position 2.877 1.757Promotion =1 if promoted this year; =0 otherwise 0.139 0.346Service Length Total service years in this city 6.740 8.004As a leader in aFirm =1 if worked as the CEO in a big enterprise; =0 otherwise 0.163 0.370Up-tier government =1 if worked in a provincial/central government; =0 otherwise 0.589 0.492County =1 if worked as a county governor/secretary; =0 otherwise 0.396 0.490Firm variablesDesulfurization Rate A fossil-fuel power plant’s desulfurization rate (%) 22.620 40.03Firm Size A fossil-fuel power plant’s electricity generation capacity (104 KW) 19.108 41.17Firm Age A fossil-fuel power plant’s age (year) 21.540 19.03Other ControlsRoughness Standard deviation of land gradient 19.989 12.044Elevation Average elevation (km) 0.521 0.593Wind Speed Annual average wind speed 1990-1995 (0.1m/s) 22.831 9.494Coldness Percent of days with temperatures of 5o or below in one year 1990-1995(%) 0.189 0.155Precipitation Annual average precipitation 1990-1995 (mm) 985.578 497.783Soil PH Average soil PH level of the topsoil (0-30 cm) 3.291 1.036Dirty Labor Total no. of employees in dirty sector in 1996 (10,000 persons) 10.496 9.776
Note: Extremely fatal accidents are officially defined as accidents with fatality over 10. ”Green” keywords include ”environmentalprotection” (huan bao or huanjing baohu) and ”pollutant emissions reduction” (jian pai). Data sources are described in full in section4.1. 27
Table 2: SO2 Intensive IndustriesSector (2-digit) SO2 emissions share in total industrial SO2 emissions
Production and Supply of Electric Power & Heat Power 50.4%Non-metallic Mineral Products 12.7%Pressing of Ferrous Metals 10.4%Raw Chemical Materials and Chemical Products 6.1%Smelting and Pressing of Non-ferrous Metals 5.8%Processing of Petroleum, Coking, Pressing of Nuclear Fuel 3.1%Sum 88.5%
Note: Data source: China Pollution Source Census 2007.
Table 3: Effect of Performance Evaluation on Industrial SO2 Emissions, PM2.5 and WastewaterDependent variable Log(SO2) Log(PM2.5) Log (Wastewater)
(1) (2) (3) (4) (5) (6)TCZ × Post2005 -0.234*** -0.145** -0.154** -0.013* -0.030
(0.061) (0.062) (0.068) (0.007) (0.056)TCZ × Year2005 -0.039
(0.050)TCZ × Post2007 -0.166**
(0.072)TCZ × Year06 07 -0.093*
(0.048)City dummy Yes Yes Yes Yes Yes YesYear dummy Yes Yes Yes Yes Yes YesControl × T Yes Yes Yes Yes YesControl × T2 Yes Yes Yes Yes YesControl × T3 Yes Yes Yes Yes YesNo. of observations 3,714 3,714 3,714 3,714 2,574 3,715R-squared 0.852 0.861 0.861 0.861 0.260 0.868Year coverage 2001-2013 2001-2013 2001-2013 2001-2013 2001-2010 2001-2013No. of clusters 286 286 286 286 286 286
Note: ***, **, * denote significance at 1%, ** at 5%, and * at 10%. All observations are at the city-level.The variable ”control” denotes seven key criteria the central government used in selecting the TCZ cities. Incolumns 1-4, the dependent variable is the natural log of industrial SO2 emissions. Interactions of the seven keyselection variables with a third-order polynomial function of time are included in columns 2-6. The PM2.5 dataobtained from the MODIS and MISR Aerosol Optical Depth are substituted for industrial SO2 emissions incolumn 5. In column 6, industrial waste water is used as the dependent variable for placebo test. All regressionscontrol for year fixed effects and city fixed effects. The standard errors are reported in parentheses, clusteredby cities.
28
Table 4: Effect of Performance Evaluation on GDP and GDP Target (2001-2013)Dependent variable GDPTarget GDPTarget Gap GDPReal Gov EcoIndex
(1) (2) (3) (4)TCZ × Post2005 -0.747*** -0.554*** -0.881*** 0.047**
(0.225) (0.208) (0.326) (0.021)City dummy Yes Yes Yes YesYear dummy Yes Yes Yes YesControl × T Yes Yes Yes YesControl × T2 Yes Yes Yes YesControl × T3 Yes Yes Yes YesNo. of observations 3,715 3,715 3,718 3,718Adjusted R-squared 0.613 0.475 0.467 0.523No. of clusters 286 286 286 286
Note: ***, **, * denote significance at 1%, ** at 5%, and * at 10%. All observationsare at the city-level. All regressions control for year fixed effects, city fixed effects, in-teractions of the seven key selection variables with a third-order polynomial functionof time. The standard errors are reported in parentheses, clustered by cities.
29
Table 5: Effect of Performance Evaluation on SO2 Intensive SectorDependent variable Log (1+CoalFirePlant Capacity) Desulfurization Rate
Entry Exit(1) (2) (3)
TCZ × Post2005 -2.012*** 0.726*** 3.873*(0.478) (0.273) (2.246)
TCZ 1.931(1.483)
Log(Firm Size) 7.719***(1.280)
Log(Firm Age) 1.931(1.483)
Firm dummy YesCity dummy Yes YesYear dummy Yes Yes YesControl × T Yes Yes YesControl × T2 Yes Yes YesControl × T3 Yes Yes YesYear coverage 2001-2010 2001-2008,2010 2001-2010
No. of observations 2,860 2,574 22,100Adjusted R-squared 0.210 0.260 0.695
No. of clusters 286 286 277Note: ***, **, * denote significance at 1%, ** at 5%, and * at 10%. In columns 1-2,observations are at the city-level. In column 3, observations are at firm-level. Allregressions control for year fixed effects, city fixed effects, interactions of the sevenkey selection variables with a third-order polynomial function of time. In column3, natural log of firm age and firm size are further added as control variables. Thestandard errors are reported in parentheses, clustered by cities.
30
Table 6: The Impact of Fatal Accidents on SO2 Emissions (2001-2013)Dependent variable Log (SO2)
Happened Death toll
(1) (2)
TCZ × Post2005× Fatal Accident -0.154* -0.027*
(0.085) (0.016)
TCZ × Post2005 -0.120* -0.184***
(0.063) (0.067)
Fatal Accident -0.076 -0.014
(0.055) (0.010)
TCZ × Fatal Accident 0.108* 0.182
(0.064) (0.012)
Fatal Accident × Post2005 0.126* 0.023*
(0.070) (0.013)
City dummy Yes Yes
Year dummy Yes Yes
Control × T Yes Yes
Control × T2 Yes Yes
Control × T3 Yes Yes
No. of observations 3,714 3,714
R-squared 0.861 0.861
No. of clusters 286 286
Note: ***, **, * denote significance at 1%, ** at 5%, and * at 10%. Allobservations are at the city-level. All regressions control for year fixedeffects, city fixed effects, interactions of the seven key selection variableswith a third-order polynomial function of time. The standard errors arereported in parentheses, clustered by cities.
31
Table 7: Political Tenure and Promotion of City Mayors and CCP Secretaries (2001-2013)Dependent Variable City Mayor CCP Secretary
New Log(Tenure) Promotion New Log(Tenure) Promotion(1) (2) (3) (4) (5) (6)
TCZ × Post2005 0.026 -0.005 0.053 0.091 -0.065 -0.109(0.073) (0.044) (0.102) (0.077) (0.049) (0.108)
City dummy Yes Yes Yes Yes Yes YesYear dummy Yes Yes Yes Yes Yes YesControl × T Yes Yes Yes Yes Yes YesControl × T2 Yes Yes Yes Yes Yes YesControl × T3 Yes Yes Yes Yes Yes YesNo. of observations 3,718 3,718 3,718 3,718 3,718 3,718Adjusted R-squared 0.067 0.059Pseudo R-squared 0.072 0.113 0.071 0.155No. of clusters 286 286 286 286 286 286
Note: All observations are at the city-level. All regressions control for year fixed effects, city fixedeffects, interactions of the seven key selection variables with a third-order polynomial function oftime. The standard errors are reported in parentheses, clustered by cities.
Table 8: Characteristics of Mayors and Secretaries (2001-2013)Dependent Variable Mayor Secretary
Log(Age) Education Hometown Log(Age) Education Hometown(1) (2) (3) (4) (5) (6)
TCZ × Post2005 0.014 0.105 0.445 -0.011 0.225 -0.004(0.009) (0.227) (0.442) (0.008) (0.214) (0.422)
City dummy Yes Yes Yes Yes Yes YesYear dummy Yes Yes Yes Yes Yes YesControl × T Yes Yes Yes Yes Yes YesControl × T2 Yes Yes Yes Yes Yes YesControl × T3 Yes Yes Yes Yes Yes YesNo. of observations 3,718 3,718 3,718 3,718 3,718 3,718Adjusted R-squared 0.291 0.364Pseudo R-squared 0.365 0.563 0.356 0.540No. of clusters 286 286 286 286 286 286
Note: All observations are at the city-level. All regressions control for year fixed effects, city fixedeffects, interactions of the seven key selection variables with a third-order polynomial function oftime. The standard errors are reported in parentheses, clustered by cities.
32
Table 9: Political and Work Experience of Mayors and Secretaries (2001-2013)Dependent variable Mayor Secretary
Log(Service As a leader in a Log(Service As a leader in aLength) Firm Up-tier County Length) Firm Up-tier County
government government(1) (2) (3) (4) (1) (2) (3) (4)
TCZ × Post2005 0.084 -0.186 -0.180 -0.069 -0.059 0.247 0.091 -0.288(0.115) (0.270) (0.191) (0.197) (0.097) (0.288) (0.203) (0.224)
City dummy Yes Yes Yes Yes Yes Yes Yes YesYear dummy Yes Yes Yes Yes Yes Yes Yes YesControl × T Yes Yes Yes Yes Yes Yes Yes YesControl × T2 Yes Yes Yes Yes Yes Yes Yes YesControl × T3 Yes Yes Yes Yes Yes Yes Yes YesNo. of observations 3,718 3,718 3,718 3,718 3,718 3,718 3,718 3,718Adjusted R-squared 0.237 0.186Pseudo R-squared 0.389 0.246 0.265 0.412 0.327 0.338No. of clusters 286 286 286 286 286 286 286 286
Note: All observations are at the city-level. All regressions control for year fixed effects, city fixed effects, interactionsof the seven key selection variables with a third-order polynomial function of time. The standard errors are reportedin parentheses, clustered by cities.
33
Figure 1: Two Control Zone
Figure 2: SO2 Emission Trend 1995-2013
Data Source: China Statistical Yearbook (various years)
34
Figure 3: Estimated Coefficients of SO2 Emissions Gap between TCZ and non-TCZ cities
Note: Each dot estimates the additional gap in emissions between TCZ and non-TCZ citiesusing the gap in 2001 as the baseline. All other control variables follow the same specificationas column 2 in Table 3
35
Figure 4: Placebo Test
Note: The figure shows the cumulative distribution of the estimated coefficients from 1,000simulations. The distribution is generated according to the placebo test discussed in section4.2
36